OpenAI cut API prices for two of its GPT-5.6 models on Thursday, lowering GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, while introducing a faster processing option for its flagship model. The price reductions mark an aggressive move in the intensifying competition among AI labs over mid-tier models used for everyday production workloads.

What Are the New GPT-5.6 Prices?

Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, down from $1.00 and $6.00 respectively. Terra falls to $2.00 and $12.00 per million tokens, down from $2.50 and $15.00. The flagship GPT-5.6 Sol maintains its pricing at $5.00 and $30.00 per million tokens, but gains a new Fast mode that runs up to 2.5 times faster at double the cost, replacing OpenAI's previous Priority Processing tier.

The price cuts apply not only to direct API usage but also to ChatGPT Work and Codex, OpenAI's coding assistant platform. These reductions make GPT-5.6 Terra roughly competitive with Anthropic's introductory Claude Sonnet 5 pricing through the end of August, intensifying the price war among frontier labs.

How Did OpenAI Fund These Price Cuts?

OpenAI attributed the price reductions partly to an unusual development: GPT-5.6 Sol itself autonomously rewrote parts of its own production serving code after launch. The changes cut serving costs by 20% and improved token-generation efficiency by more than 15%, according to the company. This self-optimization milestone represents a notable development in AI efficiency, though OpenAI's claim remains self-reported without independent verification of how much of the price cut it actually funded versus routine infrastructure improvements.

• Luna Price Reduction: 80% decrease, now the most affordable option for high-volume classification and routing tasks
• Terra Price Reduction: 20% decrease, positioning it as competitive with mid-tier alternatives from other AI labs
• Sol Fast Mode: New option offering 2.5x faster processing at double the standard price, targeting latency-sensitive applications
• Auto-Review Feature: Moving to Luna is expected to cut that feature's cost roughly tenfold

What Other Updates Did OpenAI Release This Week?

Beyond pricing changes, OpenAI updated its ChatGPT Chrome extension with new web-aware capabilities. Users can now ask questions about YouTube videos, reference open tabs in their browser, or highlight text on a page and ask follow-up questions from a side chat panel. The desktop app also now suggests URLs as users type, helping streamline web research workflows.

These updates fold in features from OpenAI's Atlas browser, which the company is discontinuing on August 9. Additionally, ImageGen inside Codex received a new lightbox and canvas interface, making it easier to explore and refine generated visuals without leaving a coding workflow.

How Should Teams Respond to These Changes?

For engineering and operations teams running production workloads on GPT-5.6 models, the near-term strategy involves shifting high-volume classification and routing tasks to Luna to capitalize on the 80% price reduction, while reserving Sol's Fast mode for the small share of requests where latency actually matters. This approach maximizes cost efficiency while maintaining performance where speed is critical.

The competitive landscape remains fluid. Anthropic's Claude Sonnet 5 introductory pricing is scheduled to rise after August 31, making today's price parity between Terra and Claude temporary by design. Industry observers are watching whether Anthropic or Google respond with their own mid-tier price cuts once Claude's introductory rate expires, and whether OpenAI publishes more detailed information on exactly how much Sol's self-optimization contributed versus other infrastructure improvements.

The broader significance of these moves reflects OpenAI's stated mission: "Making advanced intelligence more abundant and affordable is central to our mission." As the AI market matures, pricing pressure on mid-tier models is likely to intensify, potentially reshaping how enterprises allocate their AI budgets across different use cases and model tiers.